mudasir13cs/ecommerce-query-rewriting

#e-commerce-query-rewriting-dataset

Hub: mudasir13cs/ecommerce-query-rewriting

A dataset of 10,000 examples pairing ambiguous, context-dependent user queries with their fully resolved, context-aware rewrites for e-commerce product search. Built for fine-tuning LLMs to resolve pronouns, ellipsis, ordinals, and other conversational shortcuts using prior search context — the kind of resolution real shopping assistants need to handle turns like "show me that one" or "the cheaper one".

• Dataset summary
• Dataset structure
• Dataset statistics
• Query rewriting patterns
• Loading the dataset
• Training formats
• Use cases
• Quality assurance
• Citation
• License
• Acknowledgments
• Author / contact

#dataset-summary

Conversational commerce agents constantly receive queries that only make sense given prior turns — a pronoun ("it"), an ordinal ("the second one"), or a fragment ("with better camera"). This dataset provides supervised pairs of (ambiguous query, conversation context) → (unambiguous rewritten query) so models can learn to perform this resolution reliably before handing the query off to search or a product-detail lookup.

Key features

• Real-world-style e-commerce data spanning Flipkart, Amazon, and MyOnlineShop product catalogs
• Rich conversational context: previous search term, UI state, product list, and last executed command
• 7 distinct rewriting patterns covering the most common ambiguity types in shopping dialogue
• Variable product-list sizes (1–10 items) for realistic search-result scenarios
• Ready-to-use instruction-tuning format for both query rewriting and intent classification

#dataset-structure

Each example contains:

Field	Type	Description
original_query	string	Ambiguous user query requiring context resolution
context.last_search_query	string	Previous search category or term
context.current_state	string	UI state — SEARCH_RESULTS, PRODUCT_DETAIL, or INITIAL
context.product_titles	list[string]	Product names from the last search (1–10 items)
context.last_command	string	Previously executed system command
context.product_count	int	Number of products in context (1–10)
rewritten_query	string	Clear, unambiguous query with references resolved
intent	string	User intent — search, show_detail, go_back, close
command	string	System command — show_list, show_item, go_back, close

Example

{
  "original_query": "show me that one",
  "context": {
    "last_search_query": "Smartphones",
    "current_state": "SEARCH_RESULTS",
    "product_titles": [
      "iPhone 15 Pro",
      "Samsung Galaxy S24",
      "OnePlus 12",
      "Google Pixel 8",
      "Xiaomi 14",
      "Nothing Phone 2"
    ],
    "last_command": "show_list",
    "product_count": 6
  },
  "rewritten_query": "show me details of iPhone 15 Pro from Smartphones search",
  "intent": "show_detail",
  "command": "show_item"
}

#dataset-statistics

Overall

• Total examples: 10,000
• Product sources: Flipkart, Amazon, MyOnlineShop
• Format: JSON, instruction–input–output ready

Intent distribution

Intent	Count	%
show_detail	6,000	60.0%
search	3,500	35.0%
navigation	500	5.0%

Command distribution

Command	Count	%
show_item	6,000	60.0%
show_list	3,500	35.0%
go_back	500	5.0%

State distribution

State	Count	%
SEARCH_RESULTS	8,024	80.2%
PRODUCT_DETAIL	1,976	19.8%

#query-rewriting-patterns

#	Pattern	Share	Example
1	Pronoun resolution	~30%	"show me that one" → "show me details of iPhone 15 Pro from Smartphones search"
2	Ellipsis	~20%	"find cheaper" → "find Smartphones cheaper than 50000"
3	Ordinals	~15%	"show me first" → "show me details of iPhone 15 Pro from Smartphones search"
4	Product references	~15%	"tell me about iPhone" → "show me details of iPhone 15 Pro from Smartphones search"
5	Price / category	~10%	"show me under 50000" → "show me Smartphones under 50000"
6	Navigation	~5%	"go back" → "return to Smartphones search results"
7	Refinements	~5%	"higher rating" → "find Smartphones higher rating"

#loading-the-dataset

Using 🤗 Datasets (recommended)

from datasets import load_dataset

dataset = load_dataset("mudasir13cs/ecommerce-query-rewriting")

train_data = dataset["train"]
print(train_data[0])

Direct JSON loading

import json

with open("dataset_complete.json", "r") as f:
    data = json.load(f)

#training-formats

Query rewriter

{
  "instruction": "Rewrite the ambiguous query using the provided context to make it clear and searchable.",
  "input": "Context:\nPrevious search: Smartphones\nState: SEARCH_RESULTS\nProducts (6): iPhone 15 Pro, Samsung Galaxy S24, ...\n\nQuery: show me that one",
  "output": "show me details of iPhone 15 Pro from Smartphones search"
}

Intent classifier

{
  "instruction": "Classify the user's intent based on the query and context.",
  "input": "Context:\nPrevious search: Smartphones\nState: SEARCH_RESULTS\nProducts (6): iPhone 15 Pro, Samsung Galaxy S24, ...\n\nQuery: show me that one",
  "output": "show_detail"
}

#use-cases

• Fine-tuning query rewriting models — train LLMs to resolve ambiguous queries with (context, query) → rewritten_query pairs
• Intent classification — train classifiers on the intent field for dialogue-state routing
• Conversational search systems — build production agents that maintain and resolve context across turns
• Research & evaluation — benchmark rewriting approaches, study ambiguity resolution, and analyze conversational search patterns

def rewrite_query(query, context):
    # Use a fine-tuned model to resolve the query against context
    return model.predict(query, context)

#quality-assurance

Every example is validated for:

• Presence of both original_query and rewritten_query
• Complete, well-formed context object
• Rewritten query is strictly longer and more specific than the original
• Valid intent and command values
• product_count within the valid range (1–10)

#citation

If you use this dataset in your research, please cite:

@dataset{ecommerce_query_rewriting_2026,
  title={E-Commerce Query Rewriting Dataset for Context-Aware Conversational Search},
  author={Syed Mudasir},
  year={2026},
  url={https://huggingface.co/datasets/mudasir13cs/ecommerce-query-rewriting},
  note={Dataset for fine-tuning context-aware query rewriting models in e-commerce}
}

#license

Released under Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0).

• ✅ Commercial use allowed
• ✅ Modification allowed
• ✅ Distribution allowed
• ✅ Private use allowed
• ⚠️ Attribution required
• ⚠️ ShareAlike — derivatives must use the same license

#acknowledgments

• Product data sources: Flipkart, Amazon, MyOnlineShop
• The open-source ML community for tooling and frameworks
• Hugging Face for dataset hosting infrastructure

#author--contact

Mudasir — multimodal AI, VLM fine-tuning, retrieval/RAG research, and applied NLP engineering; MS AI Convergence, Soongsil University, Seoul. More projects and publications: mudasir13cs.github.io

• Hugging Face: @mudasir13cs
• GitHub: @mudasir13cs
• Email: mudasir13cs@gmail.com

---

Version: 1.0 · Last updated: July 2026